SAN JOSE, Calif.-- The old model of innovation "just
doesn't work" anymore, and EDA vendors are stepping up their own innovation to keep systems
designers more efficient and productive.
That was the message from Sanjiv Taneja, Cadence Vice President of Product
Engineering for Front End Design, who addressed the Silicon Valley Open Innovation Summit here April 2.
"It's very clear that ... the traditional innovation model is under complete
stress," he said (see slide, left).
Enabling Faster Exploration
One of the focuses of EDA today should be to provide tools to free system architects to allocate their time optimally--where appropriate and valuable--on the complex, time-consuming task of detailed micro-architectural
specifications, but leaving the vast majority of innovation at the algorithm or
system level, Taneja said.
"In this world of a disaggregated
design chain, the one thing as a system designer you control is the
architecture. It's completely under your control. You can bring all the
underlying ideas and algorithms to bear."
The key question becomes how do you enable the
system architect to explore algorithm ideas quickly, "in a time frame of
relevance," Taneja added.
There are a number of ways, both at the
implementation level and at the verification level, he argued.
At implementation, stepping up the abstraction
ladder from RTL to transaction-level modeling -- what industry analyst Gary
Smith has dubbed Electronic System Level Design -- brings enormous productivity
"You have to rise up to
the next level of abstraction...RTL to TLM and describe your function in System C
and let high-level synthesis technology help you explore the solution space in
a very comprehensive and convergent manner."
He noted those benefits were partially behind
Cadence's decision this year to acquire
Forte Design Systems and its TLM and SystemC technologies. Taneja (pictured, below right) acknowledged that while TLM has yet to hit the mainstream, it will win over
engineering teams soon because of the flexibility and agility it offers.
"Any time there is a shift (in abstraction), like from gate level to RTL
level, there are compelling benefits," Taneja said.
By writing at the TLM level, verification teams are
"abstracting away the micro architecture," writing 4X fewer lines of
code and improving verification speed and effectiveness by 5X.
After implementation, when it comes to determining
whether your design is going to work properly, challenges continue to abound.
"The verification and
validation challenge continues to rise very significantly ... because you are
integrating an unprecedented level of functionality as you move from one
process node to another."
The "old models" that Taneja described referenced
the serial fashion in which hardware and software were designed for systems,
one before the other. Today, complexity and time-to-market challenges make this
a nonstarter for the vast majority of designs.
There are a number of pre-silicon verification
approaches that can improve productivity and cut time to market, including
virtual prototyping, emulation and FPGA prototyping, each with its speed and
flexibility benefits, he said.
Taneja also delved into the productivity gains
inherent in smart IP reuse.
He cautioned that IP should fit your SoC, not the
other way around. He likened the smart IP reuse approach to an automated IP
factory that customizes and generates IP based on system requirements and
application needs resulting in a sum that is far greater than the sum of its
The last piece of the puzzle is analog-mixed signal
integration in the presence of low power, where complexity has risen quickly in
recent years. And "it's not just a few analog blocks interacting with
digital but it's a whole slew of mixed-signal blocks that have in turn their
own digital and analog pieces with complex power management schemes,"
Here it's important to work in a comprehensive
verification environment, one that leverages formal verification techniques
formerly applied only for low-power digital designs . At heart of that is
"to be able to do a macromodel of the power intent so that the proven
techniques of digital low-power verification can continue to be applied,"
--Gary Smith Webinar: “The True ESL Flow is Now Real”
--ISQED Keynote: How RTL Synthesis Must Change for Advanced Node Designs
--Q&A with Nimish Modi: Going Beyond Traditional EDA